Image Quality Assessment of Deep Learning-Based Virtual Monoenergetic Images From Single-Energy CT Pulmonary...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Imaging

Background:

• Low keV virtual monoenergetic (VME) images enhance CT angiography but typically require dual-energy CT (DECT).
• Deep learning (DL) offers a potential method to generate VME images from single-energy CT (SECT), expanding clinical accessibility.
• Previous evaluations of DL-based VME for pulmonary angiography are limited.

Purpose of the Study:

• To evaluate the objective and subjective image quality of DL-generated VME images for CT pulmonary angiography.
• To compare DL-VME images derived from SECT against standard SECT in a clinical setting.

Main Methods:

• A retrospective analysis of 52 SECT pulmonary angiography datasets was performed.
• A deep learning model (Deep-En-Chroma DL) generated 40 keV VME images from SECT data.
• Two radiologists assessed subjective image quality and vessel opacification; objective metrics included vessel contrast and contrast-to-noise ratio (CNR).

Main Results:

• DL-VME images showed significantly higher subjective image quality and vessel opacification scores than SECT (P≤0.008).
• DL-VME demonstrated superior emboli contrast (1085 vs. 331 HU, P<0.001) and improved CNR (17.8 vs. 11.1, P<0.001).
• Performance was consistent across various patient and scanner variables, with VME showing benefits for both lighter and heavier patients.

Conclusions:

• 40 keV DL-VME images derived from SECT effectively improve vessel opacification and image quality in CT pulmonary angiography.
• The image quality benefits of DL-VME over SECT are robust and consistent across diverse acquisition parameters and patient characteristics.